CN114637843A - Data processing method and device, electronic equipment and storage medium - Google Patents

Abstract

Embodiments of the present application provide a data processing method, apparatus, electronic device, and storage medium, which relate to the technical field of artificial intelligence. The data processing method includes: using a first text data training sample and a label corresponding to the first text data training sample, pre-training a text post-processing model to obtain a pre-training model; obtaining unlabeled text data and processing the unlabeled text The pseudo-label of the unlabeled text data obtained after the data is processed; based on the unlabeled text data and the pseudo-label, the pre-training model is updated to obtain the updated model; based on the second text data training samples and the second text data training samples Corresponding labels, update the updated model, and obtain the trained text post-processing model. Through the embodiments of the present application, the accuracy rate of the text post-processing model can be made higher.

Description

Data processing method, device, electronic device and storage medium

technical field

The embodiments of the present application relate to the technical field of artificial intelligence, and in particular, to a data processing method, apparatus, electronic device, and storage medium.

Background technique

Text post-processing is a process of subsequent processing of text obtained based on automatic speech recognition technology. Generally, text post-processing includes two tasks: punctuation prediction and smooth detection. A text post-processing model can be established through a multi-task learning method, thereby realizing the text post-processing operation of the text to be processed.

Currently, text post-processing models are mainly trained on supervised text data. Specifically: first obtain a large amount of standard text data from standard libraries (such as Wikipedia and other libraries) as training labels, generate training samples based on the above standard text data, and then train the samples and labels according to the above model, and perform the initial text post-processing model. Train to get the corresponding model.

In the above process, a large amount of standard text data is used for model training. However, because these standard text data usually come from a single source and involve limited application fields, the accuracy of the text post-processing model obtained by the above training method is low.

SUMMARY OF THE INVENTION

The purpose of the present application is to provide a data processing method, apparatus, electronic device and computer storage medium, which at least partially solve the above problems in the prior art.

According to a first aspect of the embodiments of the present application, a data processing method is provided, including:

Using the first text data training sample and the label corresponding to the first text data training sample, pre-training the text post-processing model to obtain a pre-training model;

Obtain the unlabeled text data and the pseudo-label of the unlabeled text data obtained by processing the unlabeled text data;

Based on unlabeled text data and pseudo-labels, the pre-training model is trained and updated, and the updated model is obtained;

Based on the second text data training samples and the labels corresponding to the second text data training samples, the updated model is trained and updated to obtain a trained text post-processing model.

According to a second aspect of the embodiments of the present application, a data processing method is provided, including:

Get the text data to be processed;

Input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first aspect.

According to a third aspect of the embodiments of the present application, a data processing method is provided, including:

Receive an instruction input through the interface of the instant messaging application for instructing to convert the input voice data into text data;

Perform text conversion on the voice data according to the instruction to obtain text data to be processed;

Input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first aspect.

According to a fourth aspect of the embodiments of the present application, a data processing method is provided, including:

receiving an instruction set through the all-in-one device input and used to instruct to convert the input voice data into text data;

Perform text conversion on the voice data according to the instruction to obtain text data to be processed;

Input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first aspect.

According to a fifth aspect of the embodiments of the present application, a data processing method is provided, including:

Receive the voice data uploaded by the public cloud client;

performing text conversion on the voice data to obtain text data to be processed;

Input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first aspect.

According to a sixth aspect of the embodiments of the present application, a data processing method is provided, including:

Receive the to-be-processed text data uploaded by the public cloud client, wherein the to-be-processed text data is obtained after the public cloud client performs text conversion on the received voice data;

Input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first aspect.

According to a seventh aspect of the embodiments of the present application, a data processing apparatus is provided. The device includes:

The model pre-training module is used for pre-training the text post-processing model by using the first text data training sample and the label corresponding to the first text data training sample to obtain a pre-training model;

The unlabeled text data and pseudo-label acquisition module is used to obtain the unlabeled text data and the pseudo-label of the unlabeled text data obtained by processing the unlabeled text data;

The first training update module is used for training and updating the pre-training model based on the unlabeled text data and pseudo-labels to obtain the updated model;

The second training and updating module is used for training and updating the updated model based on the second text data training samples and the labels corresponding to the second text data training samples to obtain a trained text post-processing model.

According to an eighth aspect of the embodiments of the present application, a data processing apparatus is provided. The device includes:

A pending text data acquisition module, used to acquire pending text data;

The first post-processing text data acquisition module is used to input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first aspect.

According to a ninth aspect of the embodiments of the present application, a data processing apparatus is provided. The device includes:

a first instruction receiving module, configured to receive an instruction input through the interface of the instant messaging application for instructing to convert the input voice data into text data;

a first text conversion module, configured to perform text conversion on the voice data according to the instruction to obtain text data to be processed;

The second post-processing text data acquisition module is configured to input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is based on the above-mentioned first aspect. method to get.

According to a tenth aspect of the embodiments of the present application, a data processing apparatus is provided. The device includes:

The second instruction receiving module is used to receive an instruction input and set through the all-in-one device and used to instruct to convert the input voice data into text data;

A second text conversion module, configured to perform text conversion on the voice data according to the instruction to obtain text data to be processed;

A third post-processing text data acquisition module, configured to input the to-be-processed text data into a text post-processing model, and acquire processed text data output by the text post-processing model; wherein, the text post-processing model is based on the above-mentioned first aspect method to get.

According to an eleventh aspect of the embodiments of the present application, a data processing apparatus is provided. The device includes:

The voice data receiving module is used to receive the voice data uploaded by the public cloud client;

A third text conversion module, configured to perform text conversion on the voice data to obtain text data to be processed;

The fourth post-processing text data acquisition module inputs the to-be-processed text data into a text post-processing model, and obtains processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first aspect above. .

According to a twelfth aspect of the embodiments of the present application, a data processing apparatus is provided. The device includes:

a to-be-processed text data receiving module, configured to receive the to-be-processed text data uploaded by the public cloud client, wherein the to-be-processed text data is obtained after the public cloud client performs text conversion on the received voice data;

A fifth post-processing text data acquisition module, configured to input the to-be-processed text data into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is based on the above-mentioned first aspect method to get.

According to a thirteenth aspect of the embodiments of the present application, there is provided an electronic device, comprising: one or more processors; a computer-readable medium configured to store one or more programs, when the one or more programs are executed by one or more A plurality of processors execute, so that one or more processors implement the data processing method according to any one of the first to sixth aspects of the above-described embodiments.

According to a fourteenth aspect of an embodiment of the present application, there is provided a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, implements any one of the first to sixth aspects of the foregoing embodiments Aspects of data processing methods.

According to a fifteenth aspect of the embodiments of the present application, a computer program is provided, which includes computer-executable instructions, and when executed, the computer-executable instructions implement any one of the first to sixth aspects of the foregoing embodiments Aspects of data processing methods.

According to the data processing solution provided by the embodiment of the present application, after the pre-training model is obtained by using the first text data training sample and the label corresponding to the first text data training sample, it is not used as the final text post-processing model. Instead, based on the unlabeled text data and its corresponding pseudo-label, the pre-training model is updated to obtain the updated model. After that, the second text data training sample and the label corresponding to the second text data training sample are used to update the updated model. After the model is updated again, the trained text post-processing model is obtained. That is to say, compared with the conventional model training process, in the solution provided by the embodiment of the present application, the pre-trained model is updated twice before the trained text post-processing model is obtained. Therefore, the text post-processing model obtained by using the data processing method provided in the embodiment of the present application has a higher accuracy rate.

Description of drawings

Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

Fig. 1a is a flow chart of the steps of the data processing method in the first embodiment of the present application;

FIG. 1b is a schematic diagram of a data processing flow according to Embodiment 1 of the present application;

2 is a flow chart of the steps of the data processing method in the second embodiment of the application;

3a is a flowchart of the steps of the data processing method in the third embodiment of the present application;

3b is a schematic diagram of a data processing flow provided according to Embodiment 3 of the present application;

4 is a flowchart of the steps of the data processing method in Embodiment 4 of the present application;

5 is a flowchart of the steps of the data processing method in Embodiment 5 of the present application;

6 is a flowchart of the steps of the data processing method in Embodiment 6 of the present application;

FIG. 7 is a flowchart of the steps of the data processing method in Embodiment 7 of the present application;

8 is a schematic structural diagram of a data processing apparatus in Embodiment 8 of the present application;

9 is a schematic structural diagram of a data processing apparatus in Embodiment 9 of the present application;

10 is a schematic structural diagram of a data processing apparatus in Embodiment 10 of the present application;

11 is a schematic structural diagram of a data processing apparatus in Embodiment 11 of the present application;

12 is a schematic structural diagram of a data processing apparatus in Embodiment 12 of the present application;

13 is a schematic structural diagram of a data processing apparatus in Embodiment 13 of the present application;

14 is a schematic structural diagram of an electronic device in Embodiment 14 of the present application;

FIG. 15 is the hardware structure of the electronic device in the fifteenth embodiment of the present application.

Detailed ways

The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only configured to explain the related invention, rather than limit the invention. In addition, it should be noted that, for the convenience of description, only the parts related to the related invention are shown in the drawings.

It should be noted that the embodiments in the present application and the features of the embodiments may be combined with each other in the case of no conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

Referring to Fig. 1a, a flow chart of steps of the data processing method according to the first embodiment of the present application is shown.

Specifically, the data processing method provided by this embodiment includes the following steps:

Step 101 , using the first text data training sample and the label corresponding to the first text data training sample to pre-train the text post-processing model to obtain a pre-training model.

In this step, the first text data training sample may be any text data obtained based on automatic speech recognition technology. For example, use the written corpus (standard text data) in the existing written corpus (standard text database) to generate a non-smooth corpus, so that the generated above-mentioned non-smooth corpus is used as the first text data training sample, and its corresponding written language The material can be the label.

The text post-processing model in this embodiment of the present invention may be any deep learning model, such as a convolutional neural network model, a cyclic neural network model, and the like. Here, the specific form of the text post-processing model is not limited.

Usually, the number of training samples is relatively large. If the labels corresponding to the training samples are obtained by manual labeling, the cost is high. Therefore, in some optional embodiments, before this step is performed, standard text data may be obtained from a standard text database, and corresponding non-smooth text data may be generated by using preset rules; the non-smooth text data may be used as the first Text data training samples, and standard text data are used as labels corresponding to the first text data training samples. Compared with the manual labeling method, the above method can reduce labor costs.

The non-smooth text data means that there are repeated words or redundant modal particles or text data with unsmooth semantics. The above-mentioned preset rules can be appropriately set by those skilled in the art according to actual needs. For example, it can be: filtering out the punctuation marks in the standard text data, and randomly adding preset modal particles (such as: "umm" in the filtered standard text data ", "ah", etc.); it can also be: filtering out the punctuation marks in the standard text data, and randomly repeating some of the words in the filtered standard text data, etc. In the embodiment of the present application, the specific content of the preset rule may be set according to the actual situation, which is not limited here.

Step 102 , acquiring unlabeled text data and pseudo-labels of the unlabeled text data obtained by processing the unlabeled text data.

In the embodiment of the present application, the pseudo-label of the unlabeled text data can be obtained by performing a text post-processing pre-operation on the unlabeled text data. Since the model has not been trained at this time, the obtained labels may not be accurate enough, so they are called pseudo labels.

In some optional embodiments, when acquiring the unlabeled text data and the pseudo-label of the unlabeled text data obtained by processing the unlabeled text data, it is possible to: obtain the unlabeled text data; Label prediction is performed on text data to obtain pseudo-labels for unlabeled text data.

Optionally, unlabeled text data can be obtained as follows:

Obtain the speech data to be recognized; use automatic speech recognition technology to recognize the speech data to be recognized, and obtain unlabeled text data.

In the above method, the speech data to be recognized is obtained first, and then the unlabeled text data is obtained by means of automatic speech recognition technology. In this way, a large amount of unlabeled text data can be quickly obtained and can be effectively applied to ASR scenarios.

Optionally, the acquisition methods of the above-mentioned label prediction model may include the following two:

The first type may be obtained after training and updating based on the pre-training model obtained in the above step 101 . Specifically, the text data training samples and their corresponding labels can be obtained again, and then based on the text data training samples and their corresponding labels obtained again, the network parameters in the pre-training model obtained in the above step 101 are trained and updated, Thus, the label prediction model is obtained.

Among them, the text data training samples obtained again may be text data to be subjected to post-processing operations that have a high degree of matching with the application field (target field) of the text post-processing model. Correspondingly, in order to ensure the accuracy of the obtained labels, you can By means of manual participation, text post-processing is performed on the re-acquired text data training samples to obtain their corresponding labels.

The second type may be obtained after training and updating based on other models with a larger scale and higher accuracy than the pre-trained model obtained in the above step 101 .

For example, an initial label prediction model can be constructed first, and the number of network layers included in the label prediction model is more than the number of network layers included in the above pre-training model, or the dimensions of each network layer in the label prediction model are larger than the above The dimensions of each network layer in the pre-training model; then, using the corpus in the existing corpus as a training sample, pre-training the above initial label prediction model to obtain a pre-trained label prediction model; then obtain and text post-processing model The application field (target field) of the text data training samples and their labels with high matching degree, the network parameters of the pre-trained label prediction model are trained and updated, and finally the trained label prediction model is obtained. Since the number of network layers of the label prediction model or the dimension of each network layer is larger, the accuracy of the label prediction model is also higher.

Comparing the acquisition methods of the above two label prediction models: the first method is based on the pre-training model obtained in step 101, and is trained and updated. Therefore, the implementation process is relatively simple, and the acquisition speed of the label prediction model is faster. ; The second method is not based on the pre-training model obtained in step 101, but based on another model with a larger scale and higher accuracy than the pre-training model obtained in the above-mentioned step 101. Therefore, the training process of the model is more complicated, and the acquisition speed of the label prediction model is slow, but the accuracy of the acquired label prediction model is higher.

Specifically, in some optional embodiments, a label prediction model is used to perform label prediction on unlabeled text data to obtain pseudo-labels of the unlabeled text data, which may include:

Based on the third text data training samples and the labels corresponding to the third text data training samples, the pre-training model is trained and updated to obtain a label prediction model; the label prediction model is used to perform label prediction on the unlabeled text data to obtain the unlabeled text Pseudo labels for the data.

The above-mentioned third text data training samples may be: text data to be subjected to post-processing operations that have a high degree of matching with the application field (target field) of the text post-processing model. Correspondingly, in order to ensure the accuracy of the obtained labels, manual In the way of participation, text post-processing is performed on the third text data training sample, and then a label corresponding to the third text data training sample is obtained.

For example, the above-mentioned third text data training samples and their labels may be: corpora in a small-scale manually-labeled spoken language corpus. Specifically, the corpus in the manually labeled spoken language corpus may be used as the third text data training sample, and the corresponding manually labeled content may be used as the label.

In other optional embodiments, a label prediction model is used to perform label prediction on unlabeled text data to obtain pseudo-labels of the unlabeled text data, which may include:

Obtain a pre-built initial label prediction model; wherein the label prediction model contains more network layers than the text post-processing model, and/or the dimensions of each network layer in the label prediction model are larger than the text The dimensions of each network layer in the post-processing model;

Using the fourth text data training sample and the label corresponding to the fourth text data training sample, pre-training the initial label prediction model to obtain the pre-trained label prediction model;

Based on the fifth text data training sample and the label corresponding to the fifth text data training sample, training and updating the pre-trained label prediction model to obtain a trained label prediction model;

The trained label prediction model is used to perform label prediction on unlabeled text data to obtain pseudo-labels of unlabeled text data.

Among them, as far as the fourth text data training sample is concerned, since its function is the same as that of the first text data training sample in the above step 101, it is used for model pre-training. Therefore, in some optional embodiments , the written corpus in the existing written corpus can also be used to generate the non-smooth corpus, so that the generated non-smooth corpus is used as the fourth text data training sample, and the corresponding written corpus can be the label.

For the fifth text data training sample, its function is the same as that of the above-mentioned third text data training sample, both of which are used to train and update the model. Therefore, in some optional embodiments, a smaller The corpus in the large-scale human-annotated spoken language corpus is used as the fifth text data training sample, and its corresponding artificially-annotated content is used as a label.

Step 103: Based on the unlabeled text data and the pseudo-label, the pre-trained model is trained and updated to obtain an updated model.

The pre-training model is obtained by pre-training based on the corpus in the existing corpus. Due to the limited application fields involved in the existing corpus, the matching degree with the application field of the text post-processing model is not high, so the pre-training model will be caused. of lower accuracy. Based on the above reasons, after obtaining the pre-training model, a large amount of unlabeled text data (such as ASR manual transcription text) involving more application fields can be obtained, and then based on the above unlabeled text data and its pseudo-labels, pre-training The network parameters of the model are trained and updated, so that an updated model with higher accuracy can be obtained.

Step 104: Based on the second text data training samples and the labels corresponding to the second text data training samples, the updated model is trained and updated to obtain a trained text post-processing model.

Similar to the above-mentioned third text data training sample and the fifth text data training sample, the second text data training sample in this step is used as a sample for model training and updating, and can be: the application field (target) of the text post-processing model. field) with a high degree of matching to the text data to be post-processed. Correspondingly, in order to ensure the accuracy of the obtained label, the fifth text data training sample can be subjected to text post-processing by manual participation, so as to obtain the same value as the first one. The labels corresponding to the five text data training samples. For example, in some optional embodiments, the corpus in a small-scale manually labeled spoken language corpus may also be used as the second text data training sample, and the corresponding manually labeled content may be used as a label.

According to the data processing method, device, electronic device, and storage medium provided by the embodiments of the present application, the data processing method is: using a first text data training sample and a label corresponding to the first text data training sample to pre-process a text post-processing model. training to obtain a pre-training model; obtaining unlabeled text data and pseudo-labels of the unlabeled text data obtained by processing the unlabeled text data; updating the pre-training model based on the unlabeled text data and pseudo-labels, and obtaining the updated A model; based on the second text data training samples and the labels corresponding to the second text data training samples, the updated model is updated to obtain a trained text post-processing model.

In the embodiment of the present application, after the pre-training model is obtained by using the first text data training sample and the label corresponding to the first text data training sample, it is not used as the final text post-processing model. Instead, based on the unlabeled text data and its corresponding pseudo-label, the pre-training model is updated to obtain the updated model. After that, the second text data training sample and the label corresponding to the second text data training sample are used to update the updated model. After the model is updated again, the trained text post-processing model is obtained. That is to say, compared with the conventional model training process, in the model training method provided by the embodiment of the present application, the pre-training model is updated twice before the trained text post-processing model is obtained. Therefore, the text post-processing model obtained by using the data processing method provided in the embodiment of the present application has a higher accuracy rate.

The data processing methods provided in the embodiments of the present application may be executed by any appropriate device with data processing capabilities, including but not limited to: terminals, mobile terminals, PCs, servers, and the like.

Referring to FIG. 1b, FIG. 1b is a schematic diagram of a data processing flow according to Embodiment 1 of the present application. The following briefly describes the data processing flow provided in Embodiment 1 of the present application with reference to FIG. 1b, which mainly includes:

After building the initial text post-processing model, e.g. after building the initial Transformer model,

Step 1: Use the first text data training samples and their corresponding labels to pre-train the constructed initial text post-processing model to obtain a pre-trained model. Specifically: the initial Transformer model can be pre-trained by using the corpus in a large-scale existing written corpus, wherein the non-smooth data can be generated from the corpus by using preset rules, as the first text data training sample, and the The corpus corresponding to the non-smooth data is used as the label.

Step 2: Obtain the third article data training samples and their corresponding samples, and train and update the above pre-training model to obtain a prediction model. Specifically, the corpus in a small-scale manually-labeled spoken language corpus may be used as the third text data training sample, and the corresponding manually-labeled content may be used as a label.

The third step: using the unlabeled text data obtained by the automatic speech recognition technology to carry out the self-training of the model, specifically: using the above prediction model to perform label prediction on the unlabeled text data obtained by the automatic speech recognition technology, and obtain the unlabeled Pseudo-labels of text data; based on unlabeled text data and pseudo-labels, the pre-training model is trained and updated to obtain the updated model;

Step 4: Based on the second text data training samples and their corresponding labels, the updated model is trained and updated to obtain a trained text post-processing model. Specifically, the corpus in the small-scale manually-labeled spoken language corpus may be used as the second text data training sample, and the corresponding manually-labeled content may be used as the label.

Referring to FIG. 2 , a flow chart of the steps of the data processing method according to the second embodiment of the present application is shown.

Specifically, the data processing method provided in the embodiment of the present application includes the following steps:

Step 201, acquiring text data to be processed.

Step 202: Input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model.

The text post-processing model may be obtained based on the data processing method in the first embodiment, and details are not described herein again.

In this embodiment of the present application, after acquiring the text data to be processed, the text data to be processed is input into the text post-processing model obtained based on the data processing method of the first embodiment, and then the processed text data output by the text post-processing model is obtained.

Because the first embodiment uses the first text data training sample and the label corresponding to the first text data training sample in the data processing process of the text post-processing model to obtain the pre-training model, it is not used as the final text post-processing model. Instead, based on the unlabeled text data and its corresponding pseudo-label, the pre-training model is updated to obtain the updated model. After that, the second text data training sample and the label corresponding to the second text data training sample are used to update the updated model. After the model is updated again, the trained text post-processing model is obtained. That is to say, compared with the conventional model training process, in the data processing method provided in the above-mentioned first embodiment, the pre-training model is updated twice before the trained text post-processing model is obtained. Therefore, the text post-processing model obtained by using the data processing method provided in the first embodiment has a higher accuracy rate.

Furthermore, by inputting the text data to be processed into the text post-processing model obtained by using the data processing method provided in the first embodiment, processed text data with higher accuracy can be obtained.

The data processing methods provided in the embodiments of the present application may be executed by any appropriate device with data processing capabilities, including but not limited to: terminals, mobile terminals, PCs, servers, and the like.

Referring to FIG. 3 a , a flowchart of steps of the data processing method in Embodiment 3 of the present application is shown.

Specifically, the data processing method provided in the embodiment of the present application includes the following steps:

Step 301: Acquire the text data of the online log reflow as the text data to be processed.

Since the text data of the online log reflow is the actual processing object of the text post-processing model in the application stage, that is to say, the text data of the online log reflow involves the application field, which is the target field of the text post-processing model application.

Therefore, take it as the text data to be processed, after the pseudo-label of the online log reflow is predicted by the text post-processing model, the text data and its pseudo-label of the online log return are used as the training sample, and the model is trained and updated (fine Adjustment), with the increase of online log reflow text data, the accuracy of the text post-processing model can be continuously improved.

Step 302: Input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model.

The text post-processing model may be obtained based on the data processing method in the first embodiment.

Input the text data of the online log reflow into the text post-processing model, and use the model to perform label prediction on the text data of the online log reflow, and obtain the pseudo-label data of the online log reflow, which is the output of the text post-processing model in this step. processed text data.

Step 303: Based on the text data to be processed and the processed text data, the updated model is trained and updated to obtain a transition model.

Wherein, the above-mentioned updated model is the updated model in the above-mentioned first embodiment.

In this step, the updated model in the above-mentioned first embodiment is fine-tuned by using the text data returned from the online log and its pseudo-tags, that is, the network parameters of the model are trained and updated to obtain a transition model.

Step 304 , based on the sixth text data training sample and the label corresponding to the sixth text data training sample, train and update the transition model to obtain a hot standby model.

The sixth text data training sample in this step is used as a sample for model training and updating, and can also be: text data to be post-processed that has a high degree of matching with the application field (target field) of the text post-processing model, corresponding to Specifically, in order to ensure the accuracy of the obtained label, the sixth text data training sample can be subjected to text post-processing by means of manual participation, so as to obtain the label corresponding to the fifth text data training sample.

For example, the corpus in a small-scale manually-labeled spoken language corpus can be used as the fifth text data training sample, and the corresponding specific text content can be used as a label.

In this step, the transition model obtained in step 303 is fine-tuned by using small-scale manually annotated spoken language corpus data, that is, the network parameters of the transition model are trained and updated to obtain a hot standby model.

In step 305, the accuracy rates of the hot standby model and the text post-processing model are calculated respectively.

In some optional embodiments, the accuracy of the hot-standby model and the text post-processing model can be calculated in the following manner:

Obtain the seventh text data training sample and the label corresponding to the seventh text data training sample; based on the seventh text data training sample and the label corresponding to the seventh text data training sample, calculate the accuracy of the hot standby model and the text post-processing model respectively. Rate.

The seventh text data training sample in this step, as a sample for model accuracy verification, may be text data to be post-processed that has a high degree of matching with the application field (target field) of the text post-processing model. Correspondingly, it is also possible to perform text post-processing on the seventh text data training sample by means of manual participation, so as to obtain a label corresponding to the seventh text data training sample. For example, the corpus in a small-scale manually labeled spoken language corpus can be used as the seventh text data training sample, and the corresponding specific text content can be used as a label.

Step 306 , when the accuracy rate of the text post-processing model is lower than that of the hot-standby model, the hot-standby model is used as a new text post-processing model for the next text post-processing operation.

Step 307 , when the accuracy rate of the hot standby model is lower than the accuracy rate of the text post-processing model, the text post-processing model is used to perform the next text post-processing operation.

In the embodiment of the present application, after acquiring the text data of the online log reflow as the text data to be processed, the text data to be processed is input into the text post-processing model obtained based on the data processing method of the first embodiment, and the text post-processing is obtained. The processed text data output by the model. Then, based on the text data to be processed and the processed text data, the updated model in the above-mentioned first embodiment is trained and updated to obtain a transition model; based on the sixth text data training sample and the label corresponding to the sixth text data training sample, The transition model is trained and updated to obtain the hot standby model; the accuracy of the hot standby model and the text post-processing model are calculated separately; when the accuracy of the text post-processing model is lower than that of the hot standby model, the hot standby model is used as the new Text post-processing model for the next text post-processing operation.

Because the first embodiment uses the first text data training sample and the label corresponding to the first text data training sample in the data processing process of the text post-processing model to obtain the pre-training model, it is not used as the final text post-processing model. Instead, based on the unlabeled text data and its corresponding pseudo-label, the pre-training model is updated to obtain the updated model. After that, the second text data training sample and the label corresponding to the second text data training sample are used to update the updated model. After the model is updated again, the trained text post-processing model is obtained. That is to say, compared with the conventional model training process, in the data processing method provided in the above-mentioned first embodiment, the pre-training model is updated twice before the trained text post-processing model is obtained. Therefore, the text post-processing model obtained by using the data processing method provided in the first embodiment has a higher accuracy rate.

Furthermore, by inputting the text data to be processed into the text post-processing model obtained by using the data processing method provided in the first embodiment, processed text data with higher accuracy can be obtained.

In addition, in the third embodiment of the present application, the text data returned from the online log is used as the text data to be processed, and then based on the text data to be processed and the corresponding processed text data, the updated model obtained in the first embodiment After training and updating, the transition model was obtained; then the transition model was trained and updated to obtain the hot standby model; the accuracy of the hot standby model and the text post-processing model were calculated respectively, and the model with high accuracy was used as the next post-processing model. The text postprocessing model to use when processing operations. Therefore, the text post-processing model can be continuously updated through the text data returned from the online log, which further improves the accuracy of the text post-processing model.

The data processing methods provided in the embodiments of the present application may be executed by any appropriate device with data processing capabilities, including but not limited to: terminals, mobile terminals, PCs, servers, and the like.

Referring to FIG. 3b, FIG. 3b is a schematic diagram of a data processing flow according to Embodiment 3 of the present application.

The following briefly describes the data processing flow provided by the third embodiment of the present application with reference to FIG. 3b, which mainly includes:

After getting the text post-processing model,

Step 1: Self-train the text post-processing model by using the text data of the online log reflow, specifically: input the unlabeled text data of the online log reflow into the text post-processing model, and obtain the pseudo-label of the online log reflow text data (text data after processing), based on the text data of the online log reflow and the pseudo-label of the online log reflow text data, the updated model (obtained in the first embodiment) is trained and updated to obtain a transition model;

Step 2: Based on the sixth text data training sample and its label, train and update the transition model to obtain a hot standby model;

Step 3: Check whether the accuracy of the hot standby model is higher than the text post-processing model in the first step. If it is higher, use the hot standby model to replace the text post-processing model in the first step as a new text post-processing. model for the next text post-processing operation.

Referring to FIG. 4 , a flowchart of the steps of the data processing method according to the fourth embodiment of the present application is shown. An application scenario of this embodiment may be: performing text conversion on the instant messaging voice data in the instant messaging application, and performing post-processing on the converted text data.

Specifically, the data processing method provided in the embodiment of the present application includes the following steps:

Step 401, receiving an instruction input through an interface of an instant messaging application for instructing to convert the input voice data into text data.

Step 402: Perform text conversion on the voice data according to the instruction to obtain text data to be processed.

Specifically, an automatic speech recognition technology may be used to perform text conversion on the input speech data, thereby obtaining the text data to be processed.

Step 403: Input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model.

The text post-processing model may be obtained based on the data processing method in the first embodiment, and details are not described herein again.

In the embodiment of the present application, after text conversion is performed on the speech data to obtain the text data to be processed, the text data to be processed is input into the text post-processing model obtained based on the data processing method of the first embodiment, and then the text post-processing model is obtained. The output processed text data.

Because the first embodiment uses the first text data training sample and the label corresponding to the first text data training sample in the data processing process of the text post-processing model to obtain the pre-training model, it is not used as the final text post-processing model. Instead, based on the unlabeled text data and its corresponding pseudo-label, the pre-training model is updated to obtain the updated model. After that, the second text data training sample and the label corresponding to the second text data training sample are used to update the updated model. After the model is updated again, the trained text post-processing model is obtained. That is to say, compared with the conventional model training process, in the data processing method provided in the above-mentioned first embodiment, the pre-training model is updated twice before the trained text post-processing model is obtained. Therefore, the text post-processing model obtained by using the data processing method provided in the first embodiment has a higher accuracy rate.

Furthermore, by inputting the text data to be processed into the text post-processing model obtained by using the data processing method provided in the first embodiment, processed text data with higher accuracy can be obtained.

Referring to FIG. 5 , a flowchart of steps of the data processing method according to Embodiment 5 of the present application is shown. An application scenario of this embodiment may be: performing text conversion on the voice data input through the all-in-one device, and performing post-processing on the converted text data.

Specifically, the data processing method provided in the embodiment of the present application includes the following steps:

Step 501: Receive an instruction input and set through the all-in-one machine device for instructing to convert the input voice data into text data.

Specifically, an automatic speech recognition technology may be used to perform text conversion on the input speech data, thereby obtaining the text data to be processed.

Step 502: Perform text conversion on the voice data according to the instruction to obtain text data to be processed.

Step 503: Input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model.

The text post-processing model may be obtained based on the data processing method in the first embodiment, and details are not described herein again.

In the embodiment of the present application, after text conversion is performed on the speech data to obtain the text data to be processed, the text data to be processed is input into the text post-processing model obtained based on the data processing method of the first embodiment, and then the text post-processing model is obtained. The output processed text data.

Because the first embodiment uses the first text data training sample and the label corresponding to the first text data training sample in the data processing process of the text post-processing model to obtain the pre-training model, it is not used as the final text post-processing model. Instead, based on the unlabeled text data and its corresponding pseudo-label, the pre-training model is updated to obtain the updated model. After that, the second text data training sample and the label corresponding to the second text data training sample are used to update the updated model. After the model is updated again, the trained text post-processing model is obtained. That is to say, compared with the conventional model training process, in the data processing method provided in the above-mentioned first embodiment, the pre-training model is updated twice before the trained text post-processing model is obtained. Therefore, the text post-processing model obtained by using the data processing method provided in the first embodiment has a higher accuracy rate.

Furthermore, by inputting the text data to be processed into the text post-processing model obtained by using the data processing method provided in the first embodiment, processed text data with higher accuracy can be obtained.

Referring to FIG. 6 , a flowchart of the steps of the data processing method according to the sixth embodiment of the present application is shown. An application scenario of this embodiment may be: a client in a public cloud uploads voice data to a cloud server, the cloud server performs text conversion, and performs post-processing on the converted text data.

Specifically, the data processing method provided in the embodiment of the present application includes the following steps:

Step 601: Receive the voice data uploaded by the public cloud client.

Step 602: Perform text conversion on the speech data to obtain text data to be processed.

Specifically, an automatic speech recognition technology may be used to perform text conversion on the received speech data, thereby obtaining the text data to be processed.

Step 603: Input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model.

The text post-processing model may be obtained based on the data processing method in the first embodiment, and details are not described herein again.

Further, after the processed text data is obtained in the cloud server, the processed text data can also be returned to the above-mentioned public cloud client.

In the embodiment of the present application, after the cloud server performs text conversion on the voice data to obtain the text data to be processed, the cloud server inputs the text data to be processed into the text post-processing model obtained based on the data processing method of the first embodiment, and further obtains the text post-processing model. Process the processed text data output by the model.

Because the first embodiment uses the first text data training sample and the label corresponding to the first text data training sample in the data processing process of the text post-processing model to obtain the pre-training model, it is not used as the final text post-processing model. Instead, based on the unlabeled text data and its corresponding pseudo-label, the pre-training model is updated to obtain the updated model. After that, the second text data training sample and the label corresponding to the second text data training sample are used to update the updated model. After the model is updated again, the trained text post-processing model is obtained. That is to say, compared with the conventional model training process, in the data processing method provided in the above-mentioned first embodiment, the pre-training model is updated twice before the trained text post-processing model is obtained. Therefore, the text post-processing model obtained by using the data processing method provided in the first embodiment has a higher accuracy rate.

Referring to FIG. 7 , a flowchart of steps of the data processing method according to Embodiment 7 of the present application is shown. The application scenario of this embodiment can still be a public cloud scenario. Specifically, the client in the public cloud can first convert the received voice data to text, and then upload the converted text data to the cloud server, and the cloud The server performs text conversion and post-processes the converted text data.

Specifically, the data processing method provided in the embodiment of the present application includes the following steps:

Step 701: Receive the to-be-processed text data uploaded by the public cloud client.

The to-be-processed text data is obtained after the public cloud client performs text conversion on the received voice data. Specifically, an automatic speech recognition technology may be used to perform text conversion on the received speech data, thereby obtaining the text data to be processed.

Step 702: Input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model.

The text post-processing model may be obtained based on the data processing method in the first embodiment, and details are not described herein again.

Further, after the processed text data is obtained in the cloud server, the processed text data can also be returned to the above-mentioned public cloud client.

In this embodiment of the present application, after receiving the text data to be processed, the cloud server inputs the text data to be processed into the text post-processing model obtained based on the data processing method in the first embodiment, and then obtains the processed post-processing model output by the text post-processing model. text data.

Because the first embodiment uses the first text data training sample and the label corresponding to the first text data training sample in the data processing process of the text post-processing model to obtain the pre-training model, it is not used as the final text post-processing model. Instead, based on the unlabeled text data and its corresponding pseudo-label, the pre-training model is updated to obtain the updated model. After that, the second text data training sample and the label corresponding to the second text data training sample are used to update the updated model. After the model is updated again, the trained text post-processing model is obtained. That is to say, compared with the conventional model training process, in the data processing method provided in the above-mentioned first embodiment, the pre-training model is updated twice before the trained text post-processing model is obtained. Therefore, the text post-processing model obtained by using the data processing method provided in the first embodiment has a higher accuracy rate.

Referring to FIG. 8 , a schematic structural diagram of a data processing apparatus in Embodiment 8 of the present application is shown.

The data processing apparatus provided by the embodiment of the present application includes:

The model pre-training module 801 is used for pre-training the text post-processing model by using the first text data training sample and the label corresponding to the first text data training sample to obtain a pre-training model;

An unlabeled text data and pseudo-label obtaining module 802 is used to obtain the unlabeled text data and the pseudo-label of the unlabeled text data obtained by processing the unlabeled text data;

The first training update module 803 is used for training and updating the pre-training model based on the unlabeled text data and the pseudo-label to obtain the updated model;

The second training updating module 804 is configured to perform training and updating of the updated model based on the second text data training samples and the labels corresponding to the second text data training samples to obtain a trained text post-processing model.

Optionally, the device in the embodiment of the present application further includes:

The standard text data and non-smooth text data acquisition module is used to obtain standard text data from the standard text database, and use preset rules to generate corresponding non-smooth text data; use the non-smooth text data as the first text data for training sample, and use the standard text data as the label corresponding to the first text data training sample.

Optionally, the unlabeled text data and pseudo-label acquisition module 802 includes:

Unlabeled text data unit, used to obtain unlabeled text data;

The pseudo-label obtaining unit is used to perform label prediction on the unlabeled text data by using the label prediction model, and obtain the pseudo-label of the unlabeled text data.

Optional, unlabeled text data unit, specifically for:

Obtain the speech data to be recognized;

Using automatic speech recognition technology, the speech data to be recognized is recognized to obtain unlabeled text data.

Optionally, pseudo-label get unit, specifically for:

Based on the third text data training sample and the label corresponding to the third text data training sample, the pre-training model is trained and updated to obtain a label prediction model;

The label prediction model is used to predict the label of the unlabeled text data, and the pseudo-label of the unlabeled text data is obtained.

Optionally, pseudo-label get unit, specifically for:

Obtain a pre-built initial label prediction model; wherein the label prediction model contains more network layers than the text post-processing model, and/or the dimensions of each network layer in the label prediction model are larger than the text The dimensions of each network layer in the post-processing model;

Using the fourth text data training sample and the label corresponding to the fourth text data training sample, pre-training the initial label prediction model to obtain the pre-trained label prediction model;

Based on the fifth text data training sample and the label corresponding to the fifth text data training sample, training and updating the pre-trained label prediction model to obtain a trained label prediction model;

The trained label prediction model is used to perform label prediction on unlabeled text data to obtain pseudo-labels of unlabeled text data.

The data processing apparatus of the embodiment of the present application is used to implement the corresponding data processing method in the foregoing first embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here. In addition, for the function implementation of each module in the data processing apparatus of this embodiment, reference may be made to the description of the corresponding part in the foregoing method embodiment 1, and details are not repeated here.

Referring to FIG. 9 , a schematic structural diagram of a data processing apparatus in Embodiment 9 of the present application is shown.

The data processing apparatus provided by the embodiment of the present application includes:

A to-be-processed text data acquisition module 901, configured to acquire to-be-processed text data;

The first processed text data acquisition module 902 is used to input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the data processing method of the first embodiment. .

Optionally, the to-be-processed text data acquisition module 901 is specifically configured to: acquire the text data of the online log reflow as the to-be-processed text data;

The device of the embodiment of the present application further includes:

The transition model obtaining module is used to train and update the updated model in the first embodiment based on the to-be-processed text data and the processed text data after obtaining the processed text data output by the text post-processing model to obtain the transition model;

The hot standby model obtaining module is used for training and updating the transition model based on the sixth text data training sample and the label corresponding to the sixth text data training sample to obtain the hot standby model;

The accuracy rate calculation module is used to calculate the accuracy rate of the hot standby model and the text post-processing model respectively;

The text post-processing model updating module is used to use the hot-standby model as a new text post-processing model for the next text post-processing operation when the accuracy of the text post-processing model is lower than that of the hot-standby model.

Optionally, the device in the embodiment of the present application further includes:

The text post-processing model retention module is used to use the text post-processing model for the next text post-processing operation when the accuracy of the hot standby model is lower than that of the text post-processing model.

Optionally, the accuracy calculation module is specifically used for:

obtaining a seventh text data training sample and a label corresponding to the seventh text data training sample;

Based on the seventh text data training sample and the label corresponding to the seventh text data training sample, the accuracy rates of the hot standby model and the text post-processing model are calculated respectively.

The data processing apparatus of the embodiment of the present application is used to implement the corresponding data processing method in the foregoing method embodiment 2 or embodiment 3, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here. In addition, for the function implementation of each module in the data processing apparatus of the embodiment of the present application, reference may be made to the description of the corresponding part in the foregoing method embodiment 2 or embodiment 3, and details are not repeated here.

Referring to FIG. 10 , a schematic structural diagram of a data processing apparatus in Embodiment 10 of the present application is shown.

The data processing apparatus provided by the embodiment of the present application includes:

The first instruction receiving module 1001 is configured to receive an instruction input through the interface of the instant messaging application and used to instruct to convert the input voice data into text data;

The first text conversion module 1002 is configured to perform text conversion on the voice data according to the instruction to obtain the text data to be processed;

The second post-processing text data acquisition module 1003 is configured to input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the data processing method of the first embodiment. .

The data processing apparatus in the embodiment of the present application is used to implement the corresponding data processing method in the foregoing method embodiment 4, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here. In addition, for the function implementation of each module in the data processing apparatus of the embodiment of the present application, reference may be made to the description of the corresponding part in the foregoing method embodiment 4, and details are not repeated here.

Referring to FIG. 11 , a schematic structural diagram of a data processing apparatus in Embodiment 11 of the present application is shown.

The data processing apparatus provided by the embodiment of the present application includes:

The second instruction receiving module 1101 is configured to receive an instruction input and set through the all-in-one device and used to instruct to convert the input voice data into text data;

The second text conversion module 1102 is configured to perform text conversion on the voice data according to the instruction to obtain the text data to be processed;

The third post-processing text data acquisition module 1103 is configured to input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the data processing method of the first embodiment. .

The data processing apparatus in the embodiment of the present application is used to implement the corresponding data processing method in the foregoing method embodiment 5, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here. In addition, for the function implementation of each module in the data processing apparatus of the embodiment of the present application, reference may be made to the description of the corresponding part in the fifth method embodiment, which is not repeated here.

Referring to FIG. 12 , a schematic structural diagram of a data processing apparatus in Embodiment 12 of the present application is shown.

The data processing apparatus provided by the embodiment of the present application includes:

A voice data receiving module 1201, configured to receive voice data uploaded by the public cloud client;

A third text conversion module 1202, configured to perform text conversion on the voice data to obtain text data to be processed;

The fourth processed text data acquisition module 1203 inputs the text data to be processed into the text post-processing model, and obtains processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the data processing method of the first embodiment.

Optionally, the apparatus in this embodiment of the present application may further include:

The first processed text data return module is used to return the processed text data to the public cloud client.

The data processing apparatus of the embodiment of the present application is used to implement the corresponding data processing method in the foregoing method embodiment 6, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here. In addition, for the function implementation of each module in the data processing apparatus of the embodiment of the present application, reference may be made to the description of the corresponding part in the foregoing method embodiment 6, and details are not repeated here.

Referring to FIG. 13 , a schematic structural diagram of a data processing apparatus in Embodiment 13 of the present application is shown.

The data processing apparatus provided by the embodiment of the present application includes:

The to-be-processed text data receiving module 1301 is configured to receive the to-be-processed text data uploaded by the public cloud client, wherein the to-be-processed text data is obtained after the public cloud client performs text conversion on the received voice data;

The fifth processed text data acquisition module 1302 is used to input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the data processing method of the first embodiment. .

Optionally, the apparatus in this embodiment of the present application may further include:

The second processed text data return module is used to return the processed text data to the public cloud client.

The data processing apparatus in the embodiment of the present application is used to implement the corresponding data processing method in the seventh method embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here. In addition, for the function implementation of each module in the data processing apparatus of the embodiment of the present application, reference may be made to the description of the corresponding part in the seventh embodiment of the method, and details are not repeated here.

14 is a schematic structural diagram of an electronic device in Embodiment 14 of the present application; the electronic device may include:

one or more processors 1401;

Computer-readable medium 1402, which can be configured to store one or more programs,

When one or more programs are executed by one or more processors, the one or more processors implement the data processing method according to any one of the first embodiment to the seventh embodiment.

FIG. 15 is the hardware structure of the electronic device in the fifteenth embodiment of the application; as shown in FIG. 15 , the hardware structure of the electronic device may include: a processor 1501, a communication interface 1502, a computer-readable medium 1503, and a communication bus 1504;

The processor 1501, the communication interface 1502, and the computer-readable medium 1503 communicate with each other through the communication bus 704;

Optionally, the communication interface 1502 can be an interface of a communication module, such as an interface of a GSM module;

The processor 1501 may be specifically configured to: use the first text data training sample and the label corresponding to the first text data training sample to pre-train the text post-processing model to obtain a pre-training model; obtain unlabeled text data and pair the The pseudo-label of the unlabeled text data obtained after processing the unlabeled text data; based on the unlabeled text data and the pseudo-label, the pre-training model is trained and updated to obtain the updated model; the training samples based on the second text data and the second The labels corresponding to the text data training samples are used to train and update the updated model, and the trained text post-processing model is obtained.

Alternatively, the processor 1501 can also be configured to: obtain the text data to be processed; input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model; wherein, the text post-processing model is based on the first embodiment. method to get.

Alternatively, the processor 1501 can also be configured to: receive an instruction input through the interface of the instant messaging application for instructing to convert the input voice data into text data; perform text conversion on the voice data according to the instruction to obtain the text data to be processed ; Input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first embodiment.

Alternatively, the processor 1501 can also be configured to: receive an instruction input and set through the all-in-one device for instructing to convert the input voice data into text data; perform text conversion on the voice data according to the instruction to obtain the text data to be processed; Input the text data to be processed into the text post-processing model, and obtain the processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first embodiment.

Alternatively, the processor 1501 may also be configured to: receive the voice data uploaded by the public cloud client;

Perform text conversion on the voice data to obtain text data to be processed; input the text data to be processed into a text post-processing model to obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of the first embodiment.

Alternatively, the processor 1501 can also be configured to: receive the text data to be processed uploaded by the public cloud client, wherein the text data to be processed is obtained after the public cloud client performs text conversion on the received voice data; The data is input into the text post-processing model, and the processed text data output by the text post-processing model is obtained; wherein, the text post-processing model is obtained based on the method of the first embodiment.

The processor 701 may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU for short), a network processor (Network Processor, NP for short), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC). ), off-the-shelf programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The methods, steps, and logic block diagrams disclosed in the embodiments of this application can be implemented or executed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The computer-readable medium 703 may be, but is not limited to, a random access storage medium (Random Access Memory, RAM), a read-only storage medium (Read Only Memory, ROM), a programmable read-only storage medium (Programmable Read-Only Memory, PROM) ), an erasable read-only storage medium (Erasable Programmable Read-Only Memory, EPROM), an electrically erasable read-only storage medium (Electric Erasable Programmable Read-Only Memory, EEPROM), and the like.

In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including program code configured to execute the method shown in the flowchart. In such an embodiment, the computer program may be downloaded and installed from a network via the communication portion, and/or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the above-mentioned functions defined in the method of the present application are performed. It should be noted that the computer-readable medium of the present application may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the above two. The computer readable medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or a combination of any of the above. More specific examples of computer readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer disks, hard disks, random access storage media (RAM), read only storage media (ROM), erasable storage media programmable read-only storage media (EPROM or flash memory), optical fiber, portable compact disk read-only storage media (CD-ROM), optical storage media devices, magnetic storage media devices, or any suitable combination of the foregoing. In this application, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, however, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code therein. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program configured for use by or in connection with an instruction execution system, apparatus, or device . Program code embodied on a computer readable medium may be transmitted using any suitable medium including, but not limited to, wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Computer program code configured to perform the operations of the present application may be written in one or more programming languages, including object-oriented programming languages—such as Java, Smalltalk, C++, but also conventional procedures, or a combination thereof programming language - such as "C" or a similar programming language. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network: including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (eg, using an Internet service provider to connect).

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more logical functions configured to implement the specified functions executable instructions. There are specific sequence relationships in the above specific embodiments, but these sequence relationships are only exemplary, and during specific implementation, these steps may be fewer, more, or the execution order may be adjusted. That is, in some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It is also noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented in dedicated hardware-based systems that perform the specified functions or operations , or can be implemented in a combination of dedicated hardware and computer instructions.

The modules involved in the embodiments of the present application may be implemented in a software manner, and may also be implemented in a hardware manner. The described module can also be set in the processor, for example, it can be described as: a processor includes a model pre-training module, an unlabeled text data and pseudo-label acquisition module, a first training update module and a second training update module. Among them, the names of these modules do not constitute a limitation of the module itself under certain circumstances. For example, the model pre-training module can also be described as "using the first text data training samples and the corresponding first text data training samples. label, pre-train the text post-processing model, and get the module of the pre-trained model".

As another aspect, the present application also provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, implements the data processing method described in any one of the foregoing Embodiments 1 to 7.

On the other hand, the present application also provides a computer program, the computer program includes computer-executable instructions, and the computer-executable instructions, when executed, implement the data processing method described in any one of the above-mentioned Embodiments 1 to 7 . In this embodiment of the present application, the computer program may include an APP, and may also include a small program or the like.

The expressions "first," "second," "first," or "second" as used in various embodiments of the present disclosure may modify various components regardless of order and/or importance, but these expressions The corresponding parts are not restricted. The above expressions are only configured for the purpose of distinguishing an element from other elements. For example, the first user equipment and the second user equipment represent different user equipments, although both are user equipments. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of the present disclosure.

When an element (eg, a first element) is referred to as being "(operatively or communicatively) coupled" or "(operatively or communicatively) coupled to" another element (eg, a second element) When an element (eg, a second element) or is "connected to" another element (eg, a second element), it should be understood that the one element is directly connected to the other element or the one element is via yet another element (eg, a third element) is indirectly connected to the other element. In contrast, it will be understood that when an element (eg, a first element) is referred to as being "directly connected" or "directly coupled" to another element (eg, a second element), no element (eg, a third element) is interposed between the two between.

The above description is only a preferred embodiment of the present application and an illustration of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solution formed by the specific combination of the above technical features, and should also cover the above technical features or Other technical solutions formed by any combination of its equivalent features. For example, a technical solution is formed by replacing the above-mentioned features with the technical features disclosed in this application (but not limited to) with similar functions.

Claims (27)

1. A data processing method, the method comprising: Using the first text data training sample and the label corresponding to the first text data training sample, pre-training the text post-processing model to obtain a pre-training model; Obtaining unlabeled text data and the pseudo-label of the unlabeled text data obtained after processing the unlabeled text data; Based on the unlabeled text data and the pseudo-label, the pre-training model is trained and updated to obtain an updated model; Based on the second text data training samples and the labels corresponding to the second text data training samples, the updated model is trained and updated to obtain a trained text post-processing model. 2. The method according to claim 1 , wherein, in the process of using the first text data training sample and the label corresponding to the first text data training sample, the text post-processing model is pre-trained to obtain a pre-training model Before, the method further includes: Obtain standard text data from a standard text database, and use preset rules to generate corresponding non-smooth text data; The non-smooth text data is used as a first text data training sample, and the standard text data is used as a label corresponding to the first text data training sample. 3. The method according to claim 1 or 2, wherein the acquisition of unlabeled text data and the pseudo-label of the unlabeled text data obtained after processing the unlabeled text data comprises: Get unlabeled text data; A label prediction model is used to perform label prediction on the unlabeled text data to obtain a pseudo-label of the unlabeled text data. 4. The method according to claim 3, wherein said obtaining unlabeled text data comprises: Obtain the speech data to be recognized; Using automatic speech recognition technology, the to-be-recognized speech data is recognized to obtain unlabeled text data. 5. The method according to claim 3, wherein said adopting a label prediction model to perform label prediction on the unlabeled text data to obtain a pseudo-label of the unlabeled text data, comprising: Based on the third text data training sample and the label corresponding to the third text data training sample, the pre-training model is trained and updated to obtain a label prediction model; Using the label prediction model, label prediction is performed on the unlabeled text data to obtain a pseudo-label of the unlabeled text data. 6. The method according to claim 3, wherein the method of using a label prediction model to perform label prediction on the unlabeled text data to obtain a pseudo-label of the unlabeled text data comprises: Obtain a pre-built initial label prediction model; wherein, the number of network layers included in the label prediction model is more than the number of network layers included in the text post-processing model, and/or, each network layer in the label prediction model The dimension of the layer is more than the dimension of each network layer in the text post-processing model; Using a fourth text data training sample and a label corresponding to the fourth text data training sample, pre-training the initial label prediction model to obtain a pre-trained label prediction model; Based on the fifth text data training sample and the label corresponding to the fifth text data training sample, training and updating the pre-trained label prediction model to obtain a trained label prediction model; Using the trained label prediction model, perform label prediction on the unlabeled text data to obtain a pseudo-label of the unlabeled text data. 7. A data processing method, the method comprising: Get the text data to be processed; Input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of any one of claims 1-6. 8. The method according to claim 7, wherein the acquiring the text data to be processed comprises: Obtain the text data of the online log reflow as the text data to be processed; After obtaining the processed text data output by the text post-processing model, the method further includes: Based on the to-be-processed text data and the processed text data, the updated model described in any one of claims 1-6 is trained and updated to obtain a transition model; Based on the sixth text data training sample and the label corresponding to the sixth text data training sample, training and updating the transition model to obtain a hot standby model; Calculate the accuracy of the hot standby model and the text post-processing model respectively; When the accuracy rate of the text post-processing model is lower than the accuracy rate of the hot-standby model, the hot-standby model is used as a new text post-processing model for the next text post-processing operation. 9. The method according to claim 8, wherein after calculating the accuracy of the hot standby model and the text post-processing model respectively, the method further comprises: When the accuracy rate of the hot standby model is lower than the accuracy rate of the text post-processing model, the text post-processing model is used to perform the next text post-processing operation. 10. The method according to claim 8 or 9, wherein the calculating the accuracy of the hot standby model and the text post-processing model respectively comprises: obtaining a seventh text data training sample and a label corresponding to the seventh text data training sample; Based on the seventh text data training sample and the label corresponding to the seventh text data training sample, the accuracy rates of the hot standby model and the text post-processing model are calculated respectively. 11. A data processing method, the method comprising: Receive an instruction input through the interface of the instant messaging application for instructing to convert the input voice data into text data; Perform text conversion on the voice data according to the instruction to obtain text data to be processed; Input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of any one of claims 1-6. 12. A data processing method, the method comprising: receiving an instruction set through the all-in-one device input and used to instruct to convert the input voice data into text data; Perform text conversion on the voice data according to the instruction to obtain text data to be processed; Input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of any one of claims 1-6. 13. A data processing method, the method comprising: Receive the voice data uploaded by the public cloud client; performing text conversion on the voice data to obtain text data to be processed; Input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of any one of claims 1-6. 14. The method of claim 13, further comprising: Return the processed text data to the public cloud client. 15. A data processing method, the method comprising: Receive the to-be-processed text data uploaded by the public cloud client, wherein the to-be-processed text data is obtained after the public cloud client performs text conversion on the received voice data; Input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is obtained based on the method of any one of claims 1-6. 16. The method of claim 15, further comprising: Return the processed text data to the public cloud client. 17. A data processing apparatus comprising: a model pre-training module, configured to pre-train the text post-processing model by using the first text data training sample and the label corresponding to the first text data training sample to obtain a pre-training model; an unlabeled text data and pseudo-label acquisition module, used for obtaining the unlabeled text data and the pseudo-label of the unlabeled text data obtained by processing the unlabeled text data; a first training update module, used for training and updating the pre-training model based on the unlabeled text data and the pseudo-label to obtain an updated model; The second training and updating module is used for training and updating the updated model based on the second text data training samples and the labels corresponding to the second text data training samples, to obtain a trained text post-processing model. 18. A data processing apparatus comprising: A pending text data acquisition module, used to acquire pending text data; The first post-processing text data acquisition module is used to input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is based on claim 1 -6 Obtained by any of the described methods. 19. A data processing apparatus comprising: a first instruction receiving module, configured to receive an instruction input through the interface of the instant messaging application for instructing to convert the input voice data into text data; a first text conversion module, configured to perform text conversion on the voice data according to the instruction to obtain text data to be processed; The second post-processing text data acquisition module is configured to input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is based on claim 1 -6 Obtained by any one of the methods. 20. A data processing apparatus, the apparatus comprising: a second instruction receiving module, configured to receive an instruction input and set through the all-in-one device for instructing to convert the input voice data into text data; A second text conversion module, configured to perform text conversion on the voice data according to the instruction to obtain text data to be processed; A third post-processing text data acquisition module, configured to input the to-be-processed text data into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is based on claim 1 -6 Obtained by any of the described methods. 21. A data processing apparatus comprising: The voice data receiving module is used to receive the voice data uploaded by the public cloud client; A third text conversion module, configured to perform text conversion on the voice data to obtain text data to be processed; The fourth post-processing text data acquisition module inputs the to-be-processed text data into a text post-processing model, and obtains processed text data output by the text post-processing model; wherein, the text post-processing model is based on claims 1-6 obtained by any of the described methods. 22. The apparatus of claim 21, further comprising: The first processed text data return module is configured to return the processed text data to the public cloud client. 23. A data processing apparatus comprising: a to-be-processed text data receiving module, configured to receive the to-be-processed text data uploaded by the public cloud client, wherein the to-be-processed text data is obtained after the public cloud client performs text conversion on the received voice data; The fifth post-processing text data acquisition module is used to input the text data to be processed into a text post-processing model, and obtain processed text data output by the text post-processing model; wherein, the text post-processing model is based on claim 1 -6 Obtained by any of the described methods. 24. The apparatus of claim 23, further comprising: The second processed text data return module is configured to return the processed text data to the public cloud client. 25. An electronic device, comprising: a processor; and a memory configured to store computer-executable instructions that, when executed, cause the processor to implement the above claims 1 to 6 The method of any one, or the method of any one of claims 7-16. 26. A storage medium, characterized in that the storage medium stores computer-executable instructions that, when executed, implement the method according to any one of claims 1 to 6, or claim 7 -16 The method of any one. 27. A computer program, characterized in that the computer program comprises computer-executable instructions which, when executed, implement the method of any one of the preceding claims 1 to 6, or claims 7- 16. The method of any one.

Images